xref: /openbmc/linux/drivers/md/bcache/sysfs.c (revision 86db9f28)
1 // SPDX-License-Identifier: GPL-2.0
2 /*
3  * bcache sysfs interfaces
4  *
5  * Copyright 2010, 2011 Kent Overstreet <kent.overstreet@gmail.com>
6  * Copyright 2012 Google, Inc.
7  */
8 
9 #include "bcache.h"
10 #include "sysfs.h"
11 #include "btree.h"
12 #include "request.h"
13 #include "writeback.h"
14 
15 #include <linux/blkdev.h>
16 #include <linux/sort.h>
17 #include <linux/sched/clock.h>
18 
19 extern bool bcache_is_reboot;
20 
21 /* Default is 0 ("writethrough") */
22 static const char * const bch_cache_modes[] = {
23 	"writethrough",
24 	"writeback",
25 	"writearound",
26 	"none",
27 	NULL
28 };
29 
30 /* Default is 0 ("auto") */
31 static const char * const bch_stop_on_failure_modes[] = {
32 	"auto",
33 	"always",
34 	NULL
35 };
36 
37 static const char * const cache_replacement_policies[] = {
38 	"lru",
39 	"fifo",
40 	"random",
41 	NULL
42 };
43 
44 static const char * const error_actions[] = {
45 	"unregister",
46 	"panic",
47 	NULL
48 };
49 
50 write_attribute(attach);
51 write_attribute(detach);
52 write_attribute(unregister);
53 write_attribute(stop);
54 write_attribute(clear_stats);
55 write_attribute(trigger_gc);
56 write_attribute(prune_cache);
57 write_attribute(flash_vol_create);
58 
59 read_attribute(bucket_size);
60 read_attribute(block_size);
61 read_attribute(nbuckets);
62 read_attribute(tree_depth);
63 read_attribute(root_usage_percent);
64 read_attribute(priority_stats);
65 read_attribute(btree_cache_size);
66 read_attribute(btree_cache_max_chain);
67 read_attribute(cache_available_percent);
68 read_attribute(written);
69 read_attribute(btree_written);
70 read_attribute(metadata_written);
71 read_attribute(active_journal_entries);
72 read_attribute(backing_dev_name);
73 read_attribute(backing_dev_uuid);
74 
75 sysfs_time_stats_attribute(btree_gc,	sec, ms);
76 sysfs_time_stats_attribute(btree_split, sec, us);
77 sysfs_time_stats_attribute(btree_sort,	ms,  us);
78 sysfs_time_stats_attribute(btree_read,	ms,  us);
79 
80 read_attribute(btree_nodes);
81 read_attribute(btree_used_percent);
82 read_attribute(average_key_size);
83 read_attribute(dirty_data);
84 read_attribute(bset_tree_stats);
85 
86 read_attribute(state);
87 read_attribute(cache_read_races);
88 read_attribute(reclaim);
89 read_attribute(reclaimed_journal_buckets);
90 read_attribute(flush_write);
91 read_attribute(writeback_keys_done);
92 read_attribute(writeback_keys_failed);
93 read_attribute(io_errors);
94 read_attribute(congested);
95 read_attribute(cutoff_writeback);
96 read_attribute(cutoff_writeback_sync);
97 rw_attribute(congested_read_threshold_us);
98 rw_attribute(congested_write_threshold_us);
99 
100 rw_attribute(sequential_cutoff);
101 rw_attribute(data_csum);
102 rw_attribute(cache_mode);
103 rw_attribute(stop_when_cache_set_failed);
104 rw_attribute(writeback_metadata);
105 rw_attribute(writeback_running);
106 rw_attribute(writeback_percent);
107 rw_attribute(writeback_delay);
108 rw_attribute(writeback_rate);
109 
110 rw_attribute(writeback_rate_update_seconds);
111 rw_attribute(writeback_rate_i_term_inverse);
112 rw_attribute(writeback_rate_p_term_inverse);
113 rw_attribute(writeback_rate_minimum);
114 read_attribute(writeback_rate_debug);
115 
116 read_attribute(stripe_size);
117 read_attribute(partial_stripes_expensive);
118 
119 rw_attribute(synchronous);
120 rw_attribute(journal_delay_ms);
121 rw_attribute(io_disable);
122 rw_attribute(discard);
123 rw_attribute(running);
124 rw_attribute(label);
125 rw_attribute(readahead);
126 rw_attribute(errors);
127 rw_attribute(io_error_limit);
128 rw_attribute(io_error_halflife);
129 rw_attribute(verify);
130 rw_attribute(bypass_torture_test);
131 rw_attribute(key_merging_disabled);
132 rw_attribute(gc_always_rewrite);
133 rw_attribute(expensive_debug_checks);
134 rw_attribute(cache_replacement_policy);
135 rw_attribute(btree_shrinker_disabled);
136 rw_attribute(copy_gc_enabled);
137 rw_attribute(gc_after_writeback);
138 rw_attribute(size);
139 
140 static ssize_t bch_snprint_string_list(char *buf,
141 				       size_t size,
142 				       const char * const list[],
143 				       size_t selected)
144 {
145 	char *out = buf;
146 	size_t i;
147 
148 	for (i = 0; list[i]; i++)
149 		out += snprintf(out, buf + size - out,
150 				i == selected ? "[%s] " : "%s ", list[i]);
151 
152 	out[-1] = '\n';
153 	return out - buf;
154 }
155 
156 SHOW(__bch_cached_dev)
157 {
158 	struct cached_dev *dc = container_of(kobj, struct cached_dev,
159 					     disk.kobj);
160 	char const *states[] = { "no cache", "clean", "dirty", "inconsistent" };
161 	int wb = dc->writeback_running;
162 
163 #define var(stat)		(dc->stat)
164 
165 	if (attr == &sysfs_cache_mode)
166 		return bch_snprint_string_list(buf, PAGE_SIZE,
167 					       bch_cache_modes,
168 					       BDEV_CACHE_MODE(&dc->sb));
169 
170 	if (attr == &sysfs_stop_when_cache_set_failed)
171 		return bch_snprint_string_list(buf, PAGE_SIZE,
172 					       bch_stop_on_failure_modes,
173 					       dc->stop_when_cache_set_failed);
174 
175 
176 	sysfs_printf(data_csum,		"%i", dc->disk.data_csum);
177 	var_printf(verify,		"%i");
178 	var_printf(bypass_torture_test,	"%i");
179 	var_printf(writeback_metadata,	"%i");
180 	var_printf(writeback_running,	"%i");
181 	var_print(writeback_delay);
182 	var_print(writeback_percent);
183 	sysfs_hprint(writeback_rate,
184 		     wb ? atomic_long_read(&dc->writeback_rate.rate) << 9 : 0);
185 	sysfs_printf(io_errors,		"%i", atomic_read(&dc->io_errors));
186 	sysfs_printf(io_error_limit,	"%i", dc->error_limit);
187 	sysfs_printf(io_disable,	"%i", dc->io_disable);
188 	var_print(writeback_rate_update_seconds);
189 	var_print(writeback_rate_i_term_inverse);
190 	var_print(writeback_rate_p_term_inverse);
191 	var_print(writeback_rate_minimum);
192 
193 	if (attr == &sysfs_writeback_rate_debug) {
194 		char rate[20];
195 		char dirty[20];
196 		char target[20];
197 		char proportional[20];
198 		char integral[20];
199 		char change[20];
200 		s64 next_io;
201 
202 		/*
203 		 * Except for dirty and target, other values should
204 		 * be 0 if writeback is not running.
205 		 */
206 		bch_hprint(rate,
207 			   wb ? atomic_long_read(&dc->writeback_rate.rate) << 9
208 			      : 0);
209 		bch_hprint(dirty, bcache_dev_sectors_dirty(&dc->disk) << 9);
210 		bch_hprint(target, dc->writeback_rate_target << 9);
211 		bch_hprint(proportional,
212 			   wb ? dc->writeback_rate_proportional << 9 : 0);
213 		bch_hprint(integral,
214 			   wb ? dc->writeback_rate_integral_scaled << 9 : 0);
215 		bch_hprint(change, wb ? dc->writeback_rate_change << 9 : 0);
216 		next_io = wb ? div64_s64(dc->writeback_rate.next-local_clock(),
217 					 NSEC_PER_MSEC) : 0;
218 
219 		return sprintf(buf,
220 			       "rate:\t\t%s/sec\n"
221 			       "dirty:\t\t%s\n"
222 			       "target:\t\t%s\n"
223 			       "proportional:\t%s\n"
224 			       "integral:\t%s\n"
225 			       "change:\t\t%s/sec\n"
226 			       "next io:\t%llims\n",
227 			       rate, dirty, target, proportional,
228 			       integral, change, next_io);
229 	}
230 
231 	sysfs_hprint(dirty_data,
232 		     bcache_dev_sectors_dirty(&dc->disk) << 9);
233 
234 	sysfs_hprint(stripe_size,	 ((uint64_t)dc->disk.stripe_size) << 9);
235 	var_printf(partial_stripes_expensive,	"%u");
236 
237 	var_hprint(sequential_cutoff);
238 	var_hprint(readahead);
239 
240 	sysfs_print(running,		atomic_read(&dc->running));
241 	sysfs_print(state,		states[BDEV_STATE(&dc->sb)]);
242 
243 	if (attr == &sysfs_label) {
244 		memcpy(buf, dc->sb.label, SB_LABEL_SIZE);
245 		buf[SB_LABEL_SIZE + 1] = '\0';
246 		strcat(buf, "\n");
247 		return strlen(buf);
248 	}
249 
250 	if (attr == &sysfs_backing_dev_name) {
251 		snprintf(buf, BDEVNAME_SIZE + 1, "%s", dc->backing_dev_name);
252 		strcat(buf, "\n");
253 		return strlen(buf);
254 	}
255 
256 	if (attr == &sysfs_backing_dev_uuid) {
257 		/* convert binary uuid into 36-byte string plus '\0' */
258 		snprintf(buf, 36+1, "%pU", dc->sb.uuid);
259 		strcat(buf, "\n");
260 		return strlen(buf);
261 	}
262 
263 #undef var
264 	return 0;
265 }
266 SHOW_LOCKED(bch_cached_dev)
267 
268 STORE(__cached_dev)
269 {
270 	struct cached_dev *dc = container_of(kobj, struct cached_dev,
271 					     disk.kobj);
272 	ssize_t v;
273 	struct cache_set *c;
274 	struct kobj_uevent_env *env;
275 
276 	/* no user space access if system is rebooting */
277 	if (bcache_is_reboot)
278 		return -EBUSY;
279 
280 #define d_strtoul(var)		sysfs_strtoul(var, dc->var)
281 #define d_strtoul_nonzero(var)	sysfs_strtoul_clamp(var, dc->var, 1, INT_MAX)
282 #define d_strtoi_h(var)		sysfs_hatoi(var, dc->var)
283 
284 	sysfs_strtoul(data_csum,	dc->disk.data_csum);
285 	d_strtoul(verify);
286 	sysfs_strtoul_bool(bypass_torture_test, dc->bypass_torture_test);
287 	sysfs_strtoul_bool(writeback_metadata, dc->writeback_metadata);
288 	sysfs_strtoul_bool(writeback_running, dc->writeback_running);
289 	sysfs_strtoul_clamp(writeback_delay, dc->writeback_delay, 0, UINT_MAX);
290 
291 	sysfs_strtoul_clamp(writeback_percent, dc->writeback_percent,
292 			    0, bch_cutoff_writeback);
293 
294 	if (attr == &sysfs_writeback_rate) {
295 		ssize_t ret;
296 		long int v = atomic_long_read(&dc->writeback_rate.rate);
297 
298 		ret = strtoul_safe_clamp(buf, v, 1, INT_MAX);
299 
300 		if (!ret) {
301 			atomic_long_set(&dc->writeback_rate.rate, v);
302 			ret = size;
303 		}
304 
305 		return ret;
306 	}
307 
308 	sysfs_strtoul_clamp(writeback_rate_update_seconds,
309 			    dc->writeback_rate_update_seconds,
310 			    1, WRITEBACK_RATE_UPDATE_SECS_MAX);
311 	sysfs_strtoul_clamp(writeback_rate_i_term_inverse,
312 			    dc->writeback_rate_i_term_inverse,
313 			    1, UINT_MAX);
314 	sysfs_strtoul_clamp(writeback_rate_p_term_inverse,
315 			    dc->writeback_rate_p_term_inverse,
316 			    1, UINT_MAX);
317 	sysfs_strtoul_clamp(writeback_rate_minimum,
318 			    dc->writeback_rate_minimum,
319 			    1, UINT_MAX);
320 
321 	sysfs_strtoul_clamp(io_error_limit, dc->error_limit, 0, INT_MAX);
322 
323 	if (attr == &sysfs_io_disable) {
324 		int v = strtoul_or_return(buf);
325 
326 		dc->io_disable = v ? 1 : 0;
327 	}
328 
329 	sysfs_strtoul_clamp(sequential_cutoff,
330 			    dc->sequential_cutoff,
331 			    0, UINT_MAX);
332 	d_strtoi_h(readahead);
333 
334 	if (attr == &sysfs_clear_stats)
335 		bch_cache_accounting_clear(&dc->accounting);
336 
337 	if (attr == &sysfs_running &&
338 	    strtoul_or_return(buf)) {
339 		v = bch_cached_dev_run(dc);
340 		if (v)
341 			return v;
342 	}
343 
344 	if (attr == &sysfs_cache_mode) {
345 		v = __sysfs_match_string(bch_cache_modes, -1, buf);
346 		if (v < 0)
347 			return v;
348 
349 		if ((unsigned int) v != BDEV_CACHE_MODE(&dc->sb)) {
350 			SET_BDEV_CACHE_MODE(&dc->sb, v);
351 			bch_write_bdev_super(dc, NULL);
352 		}
353 	}
354 
355 	if (attr == &sysfs_stop_when_cache_set_failed) {
356 		v = __sysfs_match_string(bch_stop_on_failure_modes, -1, buf);
357 		if (v < 0)
358 			return v;
359 
360 		dc->stop_when_cache_set_failed = v;
361 	}
362 
363 	if (attr == &sysfs_label) {
364 		if (size > SB_LABEL_SIZE)
365 			return -EINVAL;
366 		memcpy(dc->sb.label, buf, size);
367 		if (size < SB_LABEL_SIZE)
368 			dc->sb.label[size] = '\0';
369 		if (size && dc->sb.label[size - 1] == '\n')
370 			dc->sb.label[size - 1] = '\0';
371 		bch_write_bdev_super(dc, NULL);
372 		if (dc->disk.c) {
373 			memcpy(dc->disk.c->uuids[dc->disk.id].label,
374 			       buf, SB_LABEL_SIZE);
375 			bch_uuid_write(dc->disk.c);
376 		}
377 		env = kzalloc(sizeof(struct kobj_uevent_env), GFP_KERNEL);
378 		if (!env)
379 			return -ENOMEM;
380 		add_uevent_var(env, "DRIVER=bcache");
381 		add_uevent_var(env, "CACHED_UUID=%pU", dc->sb.uuid),
382 		add_uevent_var(env, "CACHED_LABEL=%s", buf);
383 		kobject_uevent_env(&disk_to_dev(dc->disk.disk)->kobj,
384 				   KOBJ_CHANGE,
385 				   env->envp);
386 		kfree(env);
387 	}
388 
389 	if (attr == &sysfs_attach) {
390 		uint8_t		set_uuid[16];
391 
392 		if (bch_parse_uuid(buf, set_uuid) < 16)
393 			return -EINVAL;
394 
395 		v = -ENOENT;
396 		list_for_each_entry(c, &bch_cache_sets, list) {
397 			v = bch_cached_dev_attach(dc, c, set_uuid);
398 			if (!v)
399 				return size;
400 		}
401 		if (v == -ENOENT)
402 			pr_err("Can't attach %s: cache set not found", buf);
403 		return v;
404 	}
405 
406 	if (attr == &sysfs_detach && dc->disk.c)
407 		bch_cached_dev_detach(dc);
408 
409 	if (attr == &sysfs_stop)
410 		bcache_device_stop(&dc->disk);
411 
412 	return size;
413 }
414 
415 STORE(bch_cached_dev)
416 {
417 	struct cached_dev *dc = container_of(kobj, struct cached_dev,
418 					     disk.kobj);
419 
420 	/* no user space access if system is rebooting */
421 	if (bcache_is_reboot)
422 		return -EBUSY;
423 
424 	mutex_lock(&bch_register_lock);
425 	size = __cached_dev_store(kobj, attr, buf, size);
426 
427 	if (attr == &sysfs_writeback_running) {
428 		/* dc->writeback_running changed in __cached_dev_store() */
429 		if (IS_ERR_OR_NULL(dc->writeback_thread)) {
430 			/*
431 			 * reject setting it to 1 via sysfs if writeback
432 			 * kthread is not created yet.
433 			 */
434 			if (dc->writeback_running) {
435 				dc->writeback_running = false;
436 				pr_err("%s: failed to run non-existent writeback thread",
437 						dc->disk.disk->disk_name);
438 			}
439 		} else
440 			/*
441 			 * writeback kthread will check if dc->writeback_running
442 			 * is true or false.
443 			 */
444 			bch_writeback_queue(dc);
445 	}
446 
447 	/*
448 	 * Only set BCACHE_DEV_WB_RUNNING when cached device attached to
449 	 * a cache set, otherwise it doesn't make sense.
450 	 */
451 	if (attr == &sysfs_writeback_percent)
452 		if ((dc->disk.c != NULL) &&
453 		    (!test_and_set_bit(BCACHE_DEV_WB_RUNNING, &dc->disk.flags)))
454 			schedule_delayed_work(&dc->writeback_rate_update,
455 				      dc->writeback_rate_update_seconds * HZ);
456 
457 	mutex_unlock(&bch_register_lock);
458 	return size;
459 }
460 
461 static struct attribute *bch_cached_dev_files[] = {
462 	&sysfs_attach,
463 	&sysfs_detach,
464 	&sysfs_stop,
465 #if 0
466 	&sysfs_data_csum,
467 #endif
468 	&sysfs_cache_mode,
469 	&sysfs_stop_when_cache_set_failed,
470 	&sysfs_writeback_metadata,
471 	&sysfs_writeback_running,
472 	&sysfs_writeback_delay,
473 	&sysfs_writeback_percent,
474 	&sysfs_writeback_rate,
475 	&sysfs_writeback_rate_update_seconds,
476 	&sysfs_writeback_rate_i_term_inverse,
477 	&sysfs_writeback_rate_p_term_inverse,
478 	&sysfs_writeback_rate_minimum,
479 	&sysfs_writeback_rate_debug,
480 	&sysfs_io_errors,
481 	&sysfs_io_error_limit,
482 	&sysfs_io_disable,
483 	&sysfs_dirty_data,
484 	&sysfs_stripe_size,
485 	&sysfs_partial_stripes_expensive,
486 	&sysfs_sequential_cutoff,
487 	&sysfs_clear_stats,
488 	&sysfs_running,
489 	&sysfs_state,
490 	&sysfs_label,
491 	&sysfs_readahead,
492 #ifdef CONFIG_BCACHE_DEBUG
493 	&sysfs_verify,
494 	&sysfs_bypass_torture_test,
495 #endif
496 	&sysfs_backing_dev_name,
497 	&sysfs_backing_dev_uuid,
498 	NULL
499 };
500 KTYPE(bch_cached_dev);
501 
502 SHOW(bch_flash_dev)
503 {
504 	struct bcache_device *d = container_of(kobj, struct bcache_device,
505 					       kobj);
506 	struct uuid_entry *u = &d->c->uuids[d->id];
507 
508 	sysfs_printf(data_csum,	"%i", d->data_csum);
509 	sysfs_hprint(size,	u->sectors << 9);
510 
511 	if (attr == &sysfs_label) {
512 		memcpy(buf, u->label, SB_LABEL_SIZE);
513 		buf[SB_LABEL_SIZE + 1] = '\0';
514 		strcat(buf, "\n");
515 		return strlen(buf);
516 	}
517 
518 	return 0;
519 }
520 
521 STORE(__bch_flash_dev)
522 {
523 	struct bcache_device *d = container_of(kobj, struct bcache_device,
524 					       kobj);
525 	struct uuid_entry *u = &d->c->uuids[d->id];
526 
527 	/* no user space access if system is rebooting */
528 	if (bcache_is_reboot)
529 		return -EBUSY;
530 
531 	sysfs_strtoul(data_csum,	d->data_csum);
532 
533 	if (attr == &sysfs_size) {
534 		uint64_t v;
535 
536 		strtoi_h_or_return(buf, v);
537 
538 		u->sectors = v >> 9;
539 		bch_uuid_write(d->c);
540 		set_capacity(d->disk, u->sectors);
541 	}
542 
543 	if (attr == &sysfs_label) {
544 		memcpy(u->label, buf, SB_LABEL_SIZE);
545 		bch_uuid_write(d->c);
546 	}
547 
548 	if (attr == &sysfs_unregister) {
549 		set_bit(BCACHE_DEV_DETACHING, &d->flags);
550 		bcache_device_stop(d);
551 	}
552 
553 	return size;
554 }
555 STORE_LOCKED(bch_flash_dev)
556 
557 static struct attribute *bch_flash_dev_files[] = {
558 	&sysfs_unregister,
559 #if 0
560 	&sysfs_data_csum,
561 #endif
562 	&sysfs_label,
563 	&sysfs_size,
564 	NULL
565 };
566 KTYPE(bch_flash_dev);
567 
568 struct bset_stats_op {
569 	struct btree_op op;
570 	size_t nodes;
571 	struct bset_stats stats;
572 };
573 
574 static int bch_btree_bset_stats(struct btree_op *b_op, struct btree *b)
575 {
576 	struct bset_stats_op *op = container_of(b_op, struct bset_stats_op, op);
577 
578 	op->nodes++;
579 	bch_btree_keys_stats(&b->keys, &op->stats);
580 
581 	return MAP_CONTINUE;
582 }
583 
584 static int bch_bset_print_stats(struct cache_set *c, char *buf)
585 {
586 	struct bset_stats_op op;
587 	int ret;
588 
589 	memset(&op, 0, sizeof(op));
590 	bch_btree_op_init(&op.op, -1);
591 
592 	ret = bch_btree_map_nodes(&op.op, c, &ZERO_KEY, bch_btree_bset_stats);
593 	if (ret < 0)
594 		return ret;
595 
596 	return snprintf(buf, PAGE_SIZE,
597 			"btree nodes:		%zu\n"
598 			"written sets:		%zu\n"
599 			"unwritten sets:		%zu\n"
600 			"written key bytes:	%zu\n"
601 			"unwritten key bytes:	%zu\n"
602 			"floats:			%zu\n"
603 			"failed:			%zu\n",
604 			op.nodes,
605 			op.stats.sets_written, op.stats.sets_unwritten,
606 			op.stats.bytes_written, op.stats.bytes_unwritten,
607 			op.stats.floats, op.stats.failed);
608 }
609 
610 static unsigned int bch_root_usage(struct cache_set *c)
611 {
612 	unsigned int bytes = 0;
613 	struct bkey *k;
614 	struct btree *b;
615 	struct btree_iter iter;
616 
617 	goto lock_root;
618 
619 	do {
620 		rw_unlock(false, b);
621 lock_root:
622 		b = c->root;
623 		rw_lock(false, b, b->level);
624 	} while (b != c->root);
625 
626 	for_each_key_filter(&b->keys, k, &iter, bch_ptr_bad)
627 		bytes += bkey_bytes(k);
628 
629 	rw_unlock(false, b);
630 
631 	return (bytes * 100) / btree_bytes(c);
632 }
633 
634 static size_t bch_cache_size(struct cache_set *c)
635 {
636 	size_t ret = 0;
637 	struct btree *b;
638 
639 	mutex_lock(&c->bucket_lock);
640 	list_for_each_entry(b, &c->btree_cache, list)
641 		ret += 1 << (b->keys.page_order + PAGE_SHIFT);
642 
643 	mutex_unlock(&c->bucket_lock);
644 	return ret;
645 }
646 
647 static unsigned int bch_cache_max_chain(struct cache_set *c)
648 {
649 	unsigned int ret = 0;
650 	struct hlist_head *h;
651 
652 	mutex_lock(&c->bucket_lock);
653 
654 	for (h = c->bucket_hash;
655 	     h < c->bucket_hash + (1 << BUCKET_HASH_BITS);
656 	     h++) {
657 		unsigned int i = 0;
658 		struct hlist_node *p;
659 
660 		hlist_for_each(p, h)
661 			i++;
662 
663 		ret = max(ret, i);
664 	}
665 
666 	mutex_unlock(&c->bucket_lock);
667 	return ret;
668 }
669 
670 static unsigned int bch_btree_used(struct cache_set *c)
671 {
672 	return div64_u64(c->gc_stats.key_bytes * 100,
673 			 (c->gc_stats.nodes ?: 1) * btree_bytes(c));
674 }
675 
676 static unsigned int bch_average_key_size(struct cache_set *c)
677 {
678 	return c->gc_stats.nkeys
679 		? div64_u64(c->gc_stats.data, c->gc_stats.nkeys)
680 		: 0;
681 }
682 
683 SHOW(__bch_cache_set)
684 {
685 	struct cache_set *c = container_of(kobj, struct cache_set, kobj);
686 
687 	sysfs_print(synchronous,		CACHE_SYNC(&c->sb));
688 	sysfs_print(journal_delay_ms,		c->journal_delay_ms);
689 	sysfs_hprint(bucket_size,		bucket_bytes(c));
690 	sysfs_hprint(block_size,		block_bytes(c));
691 	sysfs_print(tree_depth,			c->root->level);
692 	sysfs_print(root_usage_percent,		bch_root_usage(c));
693 
694 	sysfs_hprint(btree_cache_size,		bch_cache_size(c));
695 	sysfs_print(btree_cache_max_chain,	bch_cache_max_chain(c));
696 	sysfs_print(cache_available_percent,	100 - c->gc_stats.in_use);
697 
698 	sysfs_print_time_stats(&c->btree_gc_time,	btree_gc, sec, ms);
699 	sysfs_print_time_stats(&c->btree_split_time,	btree_split, sec, us);
700 	sysfs_print_time_stats(&c->sort.time,		btree_sort, ms, us);
701 	sysfs_print_time_stats(&c->btree_read_time,	btree_read, ms, us);
702 
703 	sysfs_print(btree_used_percent,	bch_btree_used(c));
704 	sysfs_print(btree_nodes,	c->gc_stats.nodes);
705 	sysfs_hprint(average_key_size,	bch_average_key_size(c));
706 
707 	sysfs_print(cache_read_races,
708 		    atomic_long_read(&c->cache_read_races));
709 
710 	sysfs_print(reclaim,
711 		    atomic_long_read(&c->reclaim));
712 
713 	sysfs_print(reclaimed_journal_buckets,
714 		    atomic_long_read(&c->reclaimed_journal_buckets));
715 
716 	sysfs_print(flush_write,
717 		    atomic_long_read(&c->flush_write));
718 
719 	sysfs_print(writeback_keys_done,
720 		    atomic_long_read(&c->writeback_keys_done));
721 	sysfs_print(writeback_keys_failed,
722 		    atomic_long_read(&c->writeback_keys_failed));
723 
724 	if (attr == &sysfs_errors)
725 		return bch_snprint_string_list(buf, PAGE_SIZE, error_actions,
726 					       c->on_error);
727 
728 	/* See count_io_errors for why 88 */
729 	sysfs_print(io_error_halflife,	c->error_decay * 88);
730 	sysfs_print(io_error_limit,	c->error_limit);
731 
732 	sysfs_hprint(congested,
733 		     ((uint64_t) bch_get_congested(c)) << 9);
734 	sysfs_print(congested_read_threshold_us,
735 		    c->congested_read_threshold_us);
736 	sysfs_print(congested_write_threshold_us,
737 		    c->congested_write_threshold_us);
738 
739 	sysfs_print(cutoff_writeback, bch_cutoff_writeback);
740 	sysfs_print(cutoff_writeback_sync, bch_cutoff_writeback_sync);
741 
742 	sysfs_print(active_journal_entries,	fifo_used(&c->journal.pin));
743 	sysfs_printf(verify,			"%i", c->verify);
744 	sysfs_printf(key_merging_disabled,	"%i", c->key_merging_disabled);
745 	sysfs_printf(expensive_debug_checks,
746 		     "%i", c->expensive_debug_checks);
747 	sysfs_printf(gc_always_rewrite,		"%i", c->gc_always_rewrite);
748 	sysfs_printf(btree_shrinker_disabled,	"%i", c->shrinker_disabled);
749 	sysfs_printf(copy_gc_enabled,		"%i", c->copy_gc_enabled);
750 	sysfs_printf(gc_after_writeback,	"%i", c->gc_after_writeback);
751 	sysfs_printf(io_disable,		"%i",
752 		     test_bit(CACHE_SET_IO_DISABLE, &c->flags));
753 
754 	if (attr == &sysfs_bset_tree_stats)
755 		return bch_bset_print_stats(c, buf);
756 
757 	return 0;
758 }
759 SHOW_LOCKED(bch_cache_set)
760 
761 STORE(__bch_cache_set)
762 {
763 	struct cache_set *c = container_of(kobj, struct cache_set, kobj);
764 	ssize_t v;
765 
766 	/* no user space access if system is rebooting */
767 	if (bcache_is_reboot)
768 		return -EBUSY;
769 
770 	if (attr == &sysfs_unregister)
771 		bch_cache_set_unregister(c);
772 
773 	if (attr == &sysfs_stop)
774 		bch_cache_set_stop(c);
775 
776 	if (attr == &sysfs_synchronous) {
777 		bool sync = strtoul_or_return(buf);
778 
779 		if (sync != CACHE_SYNC(&c->sb)) {
780 			SET_CACHE_SYNC(&c->sb, sync);
781 			bcache_write_super(c);
782 		}
783 	}
784 
785 	if (attr == &sysfs_flash_vol_create) {
786 		int r;
787 		uint64_t v;
788 
789 		strtoi_h_or_return(buf, v);
790 
791 		r = bch_flash_dev_create(c, v);
792 		if (r)
793 			return r;
794 	}
795 
796 	if (attr == &sysfs_clear_stats) {
797 		atomic_long_set(&c->writeback_keys_done,	0);
798 		atomic_long_set(&c->writeback_keys_failed,	0);
799 
800 		memset(&c->gc_stats, 0, sizeof(struct gc_stat));
801 		bch_cache_accounting_clear(&c->accounting);
802 	}
803 
804 	if (attr == &sysfs_trigger_gc)
805 		force_wake_up_gc(c);
806 
807 	if (attr == &sysfs_prune_cache) {
808 		struct shrink_control sc;
809 
810 		sc.gfp_mask = GFP_KERNEL;
811 		sc.nr_to_scan = strtoul_or_return(buf);
812 		c->shrink.scan_objects(&c->shrink, &sc);
813 	}
814 
815 	sysfs_strtoul_clamp(congested_read_threshold_us,
816 			    c->congested_read_threshold_us,
817 			    0, UINT_MAX);
818 	sysfs_strtoul_clamp(congested_write_threshold_us,
819 			    c->congested_write_threshold_us,
820 			    0, UINT_MAX);
821 
822 	if (attr == &sysfs_errors) {
823 		v = __sysfs_match_string(error_actions, -1, buf);
824 		if (v < 0)
825 			return v;
826 
827 		c->on_error = v;
828 	}
829 
830 	sysfs_strtoul_clamp(io_error_limit, c->error_limit, 0, UINT_MAX);
831 
832 	/* See count_io_errors() for why 88 */
833 	if (attr == &sysfs_io_error_halflife) {
834 		unsigned long v = 0;
835 		ssize_t ret;
836 
837 		ret = strtoul_safe_clamp(buf, v, 0, UINT_MAX);
838 		if (!ret) {
839 			c->error_decay = v / 88;
840 			return size;
841 		}
842 		return ret;
843 	}
844 
845 	if (attr == &sysfs_io_disable) {
846 		v = strtoul_or_return(buf);
847 		if (v) {
848 			if (test_and_set_bit(CACHE_SET_IO_DISABLE,
849 					     &c->flags))
850 				pr_warn("CACHE_SET_IO_DISABLE already set");
851 		} else {
852 			if (!test_and_clear_bit(CACHE_SET_IO_DISABLE,
853 						&c->flags))
854 				pr_warn("CACHE_SET_IO_DISABLE already cleared");
855 		}
856 	}
857 
858 	sysfs_strtoul_clamp(journal_delay_ms,
859 			    c->journal_delay_ms,
860 			    0, USHRT_MAX);
861 	sysfs_strtoul_bool(verify,		c->verify);
862 	sysfs_strtoul_bool(key_merging_disabled, c->key_merging_disabled);
863 	sysfs_strtoul(expensive_debug_checks,	c->expensive_debug_checks);
864 	sysfs_strtoul_bool(gc_always_rewrite,	c->gc_always_rewrite);
865 	sysfs_strtoul_bool(btree_shrinker_disabled, c->shrinker_disabled);
866 	sysfs_strtoul_bool(copy_gc_enabled,	c->copy_gc_enabled);
867 	/*
868 	 * write gc_after_writeback here may overwrite an already set
869 	 * BCH_DO_AUTO_GC, it doesn't matter because this flag will be
870 	 * set in next chance.
871 	 */
872 	sysfs_strtoul_clamp(gc_after_writeback, c->gc_after_writeback, 0, 1);
873 
874 	return size;
875 }
876 STORE_LOCKED(bch_cache_set)
877 
878 SHOW(bch_cache_set_internal)
879 {
880 	struct cache_set *c = container_of(kobj, struct cache_set, internal);
881 
882 	return bch_cache_set_show(&c->kobj, attr, buf);
883 }
884 
885 STORE(bch_cache_set_internal)
886 {
887 	struct cache_set *c = container_of(kobj, struct cache_set, internal);
888 
889 	/* no user space access if system is rebooting */
890 	if (bcache_is_reboot)
891 		return -EBUSY;
892 
893 	return bch_cache_set_store(&c->kobj, attr, buf, size);
894 }
895 
896 static void bch_cache_set_internal_release(struct kobject *k)
897 {
898 }
899 
900 static struct attribute *bch_cache_set_files[] = {
901 	&sysfs_unregister,
902 	&sysfs_stop,
903 	&sysfs_synchronous,
904 	&sysfs_journal_delay_ms,
905 	&sysfs_flash_vol_create,
906 
907 	&sysfs_bucket_size,
908 	&sysfs_block_size,
909 	&sysfs_tree_depth,
910 	&sysfs_root_usage_percent,
911 	&sysfs_btree_cache_size,
912 	&sysfs_cache_available_percent,
913 
914 	&sysfs_average_key_size,
915 
916 	&sysfs_errors,
917 	&sysfs_io_error_limit,
918 	&sysfs_io_error_halflife,
919 	&sysfs_congested,
920 	&sysfs_congested_read_threshold_us,
921 	&sysfs_congested_write_threshold_us,
922 	&sysfs_clear_stats,
923 	NULL
924 };
925 KTYPE(bch_cache_set);
926 
927 static struct attribute *bch_cache_set_internal_files[] = {
928 	&sysfs_active_journal_entries,
929 
930 	sysfs_time_stats_attribute_list(btree_gc, sec, ms)
931 	sysfs_time_stats_attribute_list(btree_split, sec, us)
932 	sysfs_time_stats_attribute_list(btree_sort, ms, us)
933 	sysfs_time_stats_attribute_list(btree_read, ms, us)
934 
935 	&sysfs_btree_nodes,
936 	&sysfs_btree_used_percent,
937 	&sysfs_btree_cache_max_chain,
938 
939 	&sysfs_bset_tree_stats,
940 	&sysfs_cache_read_races,
941 	&sysfs_reclaim,
942 	&sysfs_reclaimed_journal_buckets,
943 	&sysfs_flush_write,
944 	&sysfs_writeback_keys_done,
945 	&sysfs_writeback_keys_failed,
946 
947 	&sysfs_trigger_gc,
948 	&sysfs_prune_cache,
949 #ifdef CONFIG_BCACHE_DEBUG
950 	&sysfs_verify,
951 	&sysfs_key_merging_disabled,
952 	&sysfs_expensive_debug_checks,
953 #endif
954 	&sysfs_gc_always_rewrite,
955 	&sysfs_btree_shrinker_disabled,
956 	&sysfs_copy_gc_enabled,
957 	&sysfs_gc_after_writeback,
958 	&sysfs_io_disable,
959 	&sysfs_cutoff_writeback,
960 	&sysfs_cutoff_writeback_sync,
961 	NULL
962 };
963 KTYPE(bch_cache_set_internal);
964 
965 static int __bch_cache_cmp(const void *l, const void *r)
966 {
967 	cond_resched();
968 	return *((uint16_t *)r) - *((uint16_t *)l);
969 }
970 
971 SHOW(__bch_cache)
972 {
973 	struct cache *ca = container_of(kobj, struct cache, kobj);
974 
975 	sysfs_hprint(bucket_size,	bucket_bytes(ca));
976 	sysfs_hprint(block_size,	block_bytes(ca));
977 	sysfs_print(nbuckets,		ca->sb.nbuckets);
978 	sysfs_print(discard,		ca->discard);
979 	sysfs_hprint(written, atomic_long_read(&ca->sectors_written) << 9);
980 	sysfs_hprint(btree_written,
981 		     atomic_long_read(&ca->btree_sectors_written) << 9);
982 	sysfs_hprint(metadata_written,
983 		     (atomic_long_read(&ca->meta_sectors_written) +
984 		      atomic_long_read(&ca->btree_sectors_written)) << 9);
985 
986 	sysfs_print(io_errors,
987 		    atomic_read(&ca->io_errors) >> IO_ERROR_SHIFT);
988 
989 	if (attr == &sysfs_cache_replacement_policy)
990 		return bch_snprint_string_list(buf, PAGE_SIZE,
991 					       cache_replacement_policies,
992 					       CACHE_REPLACEMENT(&ca->sb));
993 
994 	if (attr == &sysfs_priority_stats) {
995 		struct bucket *b;
996 		size_t n = ca->sb.nbuckets, i;
997 		size_t unused = 0, available = 0, dirty = 0, meta = 0;
998 		uint64_t sum = 0;
999 		/* Compute 31 quantiles */
1000 		uint16_t q[31], *p, *cached;
1001 		ssize_t ret;
1002 
1003 		cached = p = vmalloc(array_size(sizeof(uint16_t),
1004 						ca->sb.nbuckets));
1005 		if (!p)
1006 			return -ENOMEM;
1007 
1008 		mutex_lock(&ca->set->bucket_lock);
1009 		for_each_bucket(b, ca) {
1010 			if (!GC_SECTORS_USED(b))
1011 				unused++;
1012 			if (GC_MARK(b) == GC_MARK_RECLAIMABLE)
1013 				available++;
1014 			if (GC_MARK(b) == GC_MARK_DIRTY)
1015 				dirty++;
1016 			if (GC_MARK(b) == GC_MARK_METADATA)
1017 				meta++;
1018 		}
1019 
1020 		for (i = ca->sb.first_bucket; i < n; i++)
1021 			p[i] = ca->buckets[i].prio;
1022 		mutex_unlock(&ca->set->bucket_lock);
1023 
1024 		sort(p, n, sizeof(uint16_t), __bch_cache_cmp, NULL);
1025 
1026 		while (n &&
1027 		       !cached[n - 1])
1028 			--n;
1029 
1030 		while (cached < p + n &&
1031 		       *cached == BTREE_PRIO)
1032 			cached++, n--;
1033 
1034 		for (i = 0; i < n; i++)
1035 			sum += INITIAL_PRIO - cached[i];
1036 
1037 		if (n)
1038 			do_div(sum, n);
1039 
1040 		for (i = 0; i < ARRAY_SIZE(q); i++)
1041 			q[i] = INITIAL_PRIO - cached[n * (i + 1) /
1042 				(ARRAY_SIZE(q) + 1)];
1043 
1044 		vfree(p);
1045 
1046 		ret = scnprintf(buf, PAGE_SIZE,
1047 				"Unused:		%zu%%\n"
1048 				"Clean:		%zu%%\n"
1049 				"Dirty:		%zu%%\n"
1050 				"Metadata:	%zu%%\n"
1051 				"Average:	%llu\n"
1052 				"Sectors per Q:	%zu\n"
1053 				"Quantiles:	[",
1054 				unused * 100 / (size_t) ca->sb.nbuckets,
1055 				available * 100 / (size_t) ca->sb.nbuckets,
1056 				dirty * 100 / (size_t) ca->sb.nbuckets,
1057 				meta * 100 / (size_t) ca->sb.nbuckets, sum,
1058 				n * ca->sb.bucket_size / (ARRAY_SIZE(q) + 1));
1059 
1060 		for (i = 0; i < ARRAY_SIZE(q); i++)
1061 			ret += scnprintf(buf + ret, PAGE_SIZE - ret,
1062 					 "%u ", q[i]);
1063 		ret--;
1064 
1065 		ret += scnprintf(buf + ret, PAGE_SIZE - ret, "]\n");
1066 
1067 		return ret;
1068 	}
1069 
1070 	return 0;
1071 }
1072 SHOW_LOCKED(bch_cache)
1073 
1074 STORE(__bch_cache)
1075 {
1076 	struct cache *ca = container_of(kobj, struct cache, kobj);
1077 	ssize_t v;
1078 
1079 	/* no user space access if system is rebooting */
1080 	if (bcache_is_reboot)
1081 		return -EBUSY;
1082 
1083 	if (attr == &sysfs_discard) {
1084 		bool v = strtoul_or_return(buf);
1085 
1086 		if (blk_queue_discard(bdev_get_queue(ca->bdev)))
1087 			ca->discard = v;
1088 
1089 		if (v != CACHE_DISCARD(&ca->sb)) {
1090 			SET_CACHE_DISCARD(&ca->sb, v);
1091 			bcache_write_super(ca->set);
1092 		}
1093 	}
1094 
1095 	if (attr == &sysfs_cache_replacement_policy) {
1096 		v = __sysfs_match_string(cache_replacement_policies, -1, buf);
1097 		if (v < 0)
1098 			return v;
1099 
1100 		if ((unsigned int) v != CACHE_REPLACEMENT(&ca->sb)) {
1101 			mutex_lock(&ca->set->bucket_lock);
1102 			SET_CACHE_REPLACEMENT(&ca->sb, v);
1103 			mutex_unlock(&ca->set->bucket_lock);
1104 
1105 			bcache_write_super(ca->set);
1106 		}
1107 	}
1108 
1109 	if (attr == &sysfs_clear_stats) {
1110 		atomic_long_set(&ca->sectors_written, 0);
1111 		atomic_long_set(&ca->btree_sectors_written, 0);
1112 		atomic_long_set(&ca->meta_sectors_written, 0);
1113 		atomic_set(&ca->io_count, 0);
1114 		atomic_set(&ca->io_errors, 0);
1115 	}
1116 
1117 	return size;
1118 }
1119 STORE_LOCKED(bch_cache)
1120 
1121 static struct attribute *bch_cache_files[] = {
1122 	&sysfs_bucket_size,
1123 	&sysfs_block_size,
1124 	&sysfs_nbuckets,
1125 	&sysfs_priority_stats,
1126 	&sysfs_discard,
1127 	&sysfs_written,
1128 	&sysfs_btree_written,
1129 	&sysfs_metadata_written,
1130 	&sysfs_io_errors,
1131 	&sysfs_clear_stats,
1132 	&sysfs_cache_replacement_policy,
1133 	NULL
1134 };
1135 KTYPE(bch_cache);
1136